IMPORTING LIBRARIES¶
In [870]:
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
import seaborn as sns
from sklearn.preprocessing import Imputer
from sklearn.model_selection import train_test_split
import warnings
In [871]:
warnings.filterwarnings('ignore')
LOADING THE DATASET¶
In [872]:
vehicle = pd.read_csv("C:/Users/HP/Downloads/vehicle-1.csv")
vehicle
Out[872]:
In [873]:
#column description
vehicle.columns
Out[873]:
In [874]:
vehicle.head(10)
Out[874]:
In [875]:
vehicle.shape
Out[875]:
There are 19 columns and 846 rows in the vehicleset¶
ATTRIBUTE DATATYPES¶
In [876]:
vehicle.dtypes
Out[876]:
In [877]:
#5 number summary
vehicle.describe().transpose()
Out[877]:
CHECKING THE PRESENCE OF MISSING VALUES¶
In [878]:
vehicle.isnull().sum()
Out[878]:
There are missing values present in the vehicleset¶
So we replace it with the median of that particular column¶
In [879]:
vehicle['circularity'].fillna(vehicle['circularity'].median(),inplace=True)
vehicle['distance_circularity'].fillna(vehicle['distance_circularity'].median(),inplace=True)
vehicle['radius_ratio'].fillna(vehicle['radius_ratio'].median(),inplace=True)
vehicle['pr.axis_aspect_ratio'].fillna(vehicle['pr.axis_aspect_ratio'].median(),inplace=True)
vehicle['scatter_ratio'].fillna(vehicle['scatter_ratio'].median(),inplace=True)
vehicle['elongatedness'].fillna(vehicle['elongatedness'].median(),inplace=True)
vehicle['pr.axis_rectangularity'].fillna(vehicle['pr.axis_rectangularity'].median(),inplace=True)
vehicle['skewness_about.2'].fillna(vehicle['skewness_about.2'].median(),inplace=True)
vehicle['skewness_about.1'].fillna(vehicle['skewness_about.1'].median(),inplace=True)
vehicle['skewness_about'].fillna(vehicle['skewness_about'].median(),inplace=True)
vehicle['scaled_radius_of_gyration.1'].fillna(vehicle['scaled_radius_of_gyration.1'].median(),inplace=True)
vehicle['scaled_radius_of_gyration'].fillna(vehicle['scaled_radius_of_gyration'].median(),inplace=True)
vehicle['scaled_variance'].fillna(vehicle['scaled_variance'].median(),inplace=True)
vehicle['scaled_variance.1'].fillna(vehicle['scaled_variance.1'].median(),inplace=True)
In [880]:
vehicle.isnull().sum()
Out[880]:
There are no missing values present now.¶
CHECKING THE PRESENCE OF THE OUTLIERS IN THE DATASET.¶
In [881]:
sns.boxplot(vehicle['compactness'])
Out[881]:
In [882]:
sns.boxplot(vehicle['circularity'])
Out[882]:
In [883]:
sns.boxplot(vehicle['distance_circularity'])
Out[883]:
In [884]:
sns.boxplot(vehicle['radius_ratio'])
Out[884]:
In [885]:
sns.boxplot(vehicle['pr.axis_aspect_ratio'])
Out[885]:
In [886]:
sns.boxplot(vehicle['max.length_aspect_ratio'])
Out[886]:
In [887]:
sns.boxplot(vehicle['scatter_ratio'])
Out[887]:
In [888]:
sns.boxplot(vehicle['elongatedness'])
Out[888]:
In [889]:
sns.boxplot(vehicle['pr.axis_rectangularity'])
Out[889]:
In [890]:
sns.boxplot(vehicle['max.length_rectangularity'])
Out[890]:
In [891]:
sns.boxplot(vehicle['scaled_variance'])
Out[891]:
In [892]:
sns.boxplot(vehicle['scaled_variance.1'])
Out[892]:
In [893]:
sns.boxplot(vehicle['scaled_radius_of_gyration'])
Out[893]:
In [894]:
sns.boxplot(vehicle['scaled_radius_of_gyration.1'])
Out[894]:
In [895]:
sns.boxplot(vehicle['skewness_about'])
Out[895]:
In [896]:
sns.boxplot(vehicle['skewness_about.1'])
Out[896]:
In [897]:
sns.boxplot(vehicle['skewness_about.2'])
Out[897]:
In [898]:
sns.boxplot(vehicle['hollows_ratio'])
Out[898]:
Very few columns that are 'skewness_about','radius_ratio','pr.axis_aspect_ratio','max.length_aspect_ratio' and 'scaled_radius_of_gyration.1' have outliers '¶
Replacing the outliers with median.¶
In [901]:
import random as r
d = [r.random()*1000 for i in range(0,100)]
df = pd.DataFrame({'skewness_about': d})
median = df['skewness_about'].median()
std = df['skewness_about'].std()
outliers = (df['skewness_about'] - median).abs() > std
df[outliers] = np.nan
df['skewness_about'].fillna(median, inplace=True)
In [902]:
import random as r
d = [r.random()*1000 for i in range(0,100)]
df = pd.DataFrame({'radius_ratio': d})
median = df['radius_ratio'].median()
std = df['radius_ratio'].std()
outliers = (df['radius_ratio'] - median).abs() > std
df[outliers] = np.nan
df['radius_ratio'].fillna(median, inplace=True)
In [903]:
import random as r
d = [r.random()*1000 for i in range(0,100)]
df = pd.DataFrame({'pr.axis_aspect_ratio': d})
median = df['pr.axis_aspect_ratio'].median()
std = df['pr.axis_aspect_ratio'].std()
outliers = (df['pr.axis_aspect_ratio'] - median).abs() > std
df[outliers] = np.nan
df['pr.axis_aspect_ratio'].fillna(median, inplace=True)
In [904]:
import random as r
d = [r.random()*1000 for i in range(0,100)]
df = pd.DataFrame({'max.length_aspect_ratio': d})
median = df['max.length_aspect_ratio'].median()
std = df['max.length_aspect_ratio'].std()
outliers = (df['max.length_aspect_ratio'] - median).abs() > std
df[outliers] = np.nan
df['max.length_aspect_ratio'].fillna(median, inplace=True)
In [905]:
import random as r
d = [r.random()*1000 for i in range(0,100)]
df = pd.DataFrame({'scaled_radius_of_gyration.1': d})
median = df['scaled_radius_of_gyration.1'].median()
std = df['scaled_radius_of_gyration.1'].std()
outliers = (df['scaled_radius_of_gyration.1'] - median).abs() > std
df[outliers] = np.nan
df['scaled_radius_of_gyration.1'].fillna(median, inplace=True)
Checking the distribution of the dependent variable: class column¶
In [906]:
vehicle["class"].value_counts(normalize=True)
Out[906]:
Visualizing the distribution of target column¶
In [907]:
pd.value_counts(vehicle["class"]).plot(kind="bar")
Out[907]:
we can clearly see that the number of cars are relatively higher in number(around 51%).The rest 50 is equally divided into the category bus and van(26 and 23 respecively)¶
In [908]:
vehicle.groupby(["class"]).count()
Out[908]:
PAIRPLOT VISULATION TO SEE THE RELATIONSHIP BETWEEN VARIOUS ATTRIBUTES¶
In [909]:
# independant variables
X = vehicle.drop(['class'], axis=1)
# the dependent variable
y = vehicle[['class']]
sns.pairplot(X, diag_kind='kde')
Out[909]:
Some attributes have a strong postive relationship with each other as we can see in the graph.#In addition there are few variables which generate a very distributed scatter plot which means there is no significant relationship at all.¶
HEATMAP TO SEE THE CORRELATION BETWEEN ATTRIBUTES.¶
In [910]:
plt.figure(figsize=(15,15))
sns.heatmap(X.corr(),annot=True)
Out[910]:
The last five columns do not seem to have a strong positive correlation between them though hollows ratio and skweness about 2 have correrelation of 0.89.¶
The other attributes have good correrelation with each other as we can see in the heat map¶
STANDARDISING THE DATA USING ZSCORE¶
In [911]:
df = vehicle.drop('class', axis=1)
In [912]:
from scipy.stats import zscore
z= df.apply(zscore)
In [913]:
z.head()
Out[913]:
SPLITTING THE DATA INTO TRAIN AND TEST IN THE RATIO OF 70 AND 30.¶
In [952]:
from sklearn.model_selection import train_test_split
X,y = np.array(vehicle)[ :, 0:18], np.array(vehicle['class'])[:]
x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=1)
In [915]:
#Checking the split of vehicle
print("{0:0.2f}% vehicle is in training set".format((len(x_train)/len(vehicle.index)) * 100))
print("{0:0.2f}% vehicle is in test set".format((len(x_test)/len(vehicle.index)) * 100))
IMPLEMENTATING THE SUPPORT VECTOR MACHINE¶
In [925]:
from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
vehicle['class'] = le.fit_transform(vehicle['class'])
In [926]:
from sklearn import svm
clr = svm.SVC()
clr.fit(x_train , y_train)
Out[926]:
In [927]:
def getAccuracy(testSet, predictions):
correct = 0
for x in range(len(testSet)):
if testSet[x]== predictions[x]:
correct += 1
return (correct/float(len(testSet))) * 100.0
In [928]:
y_pred = clr.predict(x_test)
In [929]:
getAccuracy(y_test, y_pred)
Out[929]:
IMPLEMENTATING THE KFOLD CROSS VALIDATION¶
In [951]:
from sklearn.model_selection import KFold
from sklearn.model_selection import cross_val_score
from sklearn.tree import DecisionTreeRegressor
array = vehicle.values
X2 = array[:,0:18]
Y2 = array[:,18]
X_train, X_test, Y_train, Y_test = train_test_split(X2, Y2, test_size=0.50, random_state=1)
num_folds = 10
seed = 10
kfold = KFold(n_splits=num_folds, random_state=seed)
model = DecisionTreeRegressor()
results = cross_val_score(model, X2, Y2, cv=kfold)
print(results)
print("Accuracy: %.3f%% (%.3f%%)" % (results.mean()*100.0, results.std()*100.0))
IMPLEMENTING THE PCA MODULE.¶
In [931]:
cov = np.cov(z,rowvar=False)
print(cov)
In [932]:
from sklearn.decomposition import PCA
pca = PCA(n_components=18)
pca.fit(z)
Out[932]:
In [933]:
print(pca.explained_variance_)
In [934]:
print(pca.components_)
In [935]:
print(pca.explained_variance_ratio_)
In [936]:
plt.bar(list(range(0,18)),pca.explained_variance_ratio_,alpha=0.5, align='center')
plt.ylabel('Variation explained')
plt.xlabel('eigen Value')
plt.show()
In [937]:
plt.step(list(range(0,18)),np.cumsum(pca.explained_variance_ratio_), where='mid')
plt.ylabel('Cum of variation explained')
plt.xlabel('eigen Value')
plt.show()
we can see there is a sharp elbow around x=6 so we take n_components =6¶
DIMENSIONALITY REDUCTION.¶
In [938]:
pca2 = PCA(n_components=6)
pca2.fit(z)
print(pca2.components_)
print(pca2.explained_variance_ratio_)
Xpca = pca2.transform(z)
In [939]:
Xpca
Out[939]:
In [954]:
sns.pairplot(pd.DataFrame(Xpca))
Out[954]:
REPEATING 3,4,5 STEPS USING PCA DATA.¶
SPLITTING PCA DATA INTO TRAIN AND TEST SET¶
In [941]:
X,y = np.array(Xpca)[ :, 0:6], np.array(vehicle['class'])[:] #Note that the X variable contains PCA data
x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=1)
In [942]:
print("{0:0.2f}% vehicle is in training set".format((len(x_train)/len(vehicle.index)) * 100))
print("{0:0.2f}% vehicle is in test set".format((len(x_test)/len(vehicle.index)) * 100))
IMPLEMENTING SVM ON THE PCA DATA.¶
In [943]:
from sklearn import svm
clf = svm.SVC ()
clf.fit(x_train , y_train)
Out[943]:
In [944]:
def getAccuracy(testSet, predictions):
correct = 0
for x in range(len(testSet)):
if testSet[x]== predictions[x]:
correct += 1
return (correct/float(len(testSet))) * 100.0
In [945]:
y_pred = clf.predict(x_test)
In [946]:
getAccuracy(y_test , y_pred)
Out[946]:
In [947]:
from sklearn import metrics
import seaborn as sns
cm=metrics.confusion_matrix(y_test, y_pred)
df_cm = pd.DataFrame(cm)
plt.figure(figsize = (5,5))
sns.heatmap(df_cm, annot=True ,fmt='g')
Out[947]:
IMPLEMENTING KFOLD CROSS VALIDATION ON THE PCA DATA.¶
In [953]:
from sklearn.model_selection import KFold
from sklearn.model_selection import cross_val_score
from sklearn.tree import DecisionTreeRegressor
num_folds = 10
seed = 7
kfold = KFold(n_splits=num_folds, random_state=seed)
regression_model = DecisionTreeRegressor()
results = cross_val_score(regression_model, X, y, cv=kfold)
print(results)
print("Accuracy: %.3f%% (%.3f%%)" % (results.mean()*100.0, results.std()*100.0))
As we can see the we can improve the accuracy of both SVM and Kfold has increased after performing pca.¶
In [ ]:
In [ ]: